Background
Lyrebird is a discovery-stage biotech focused on fibrotic disease. They had a validated target and an in-house chemistry team, but the existing discovery pipeline was a stack of incumbent computational-chemistry suites, an in-house ADMET model and a vendor-locked ELN — with hand-off steps between each. The wet-lab was automated by a closed scheduler that didn't talk to any of them.
The challenge
From hypothesis to first measured hit was averaging four to six months at Lyrebird. The bottleneck was not chemistry or biology — it was the email steps between them. They wanted a single platform that would propose, plan, run and read out within the same loop.
What we did
- Aether for drug discovery ran the in-silico side: generative chemistry with synthetic-accessibility constraints, FEP scoring on the top series, full ADMET battery on hits.
- Aether for autonomous labs bound reagents, scheduled plates and drove the liquid-handling and acoustic-dispensing bench. Chain-of-custody and eBR were captured by the runtime.
- Active-learning agent picked the next experiment with cost-aware information gain — not the highest model score.
- Forward-deployed engineer joined the medicinal-chemistry team standing meetings for the first six weeks.
“It's the first platform we've used where the readout actually goes back into the model. That's the loop everyone talks about and nobody ships.”
The outcome
- From cold start, the loop produced a measured hit within nine weeks.
- Chain-of-custody record satisfied their GxP auditor on the first pass — no manual eBR remediation needed.
- Active-learning agent picked twelve compounds for synthesis; three of them measured as primary hits, four as secondary leads.
- Programme is now in lead optimisation; the platform handles four parallel chemistry series on the same bench footprint.